Face recognition method and system for personal identification and authentication

Abstract

The present invention comprises capturing an image of a subject to be authenticated; a step of face verification; followed by the process steps of a scan line detection test, a specular reflection detection test, and a chromatic moment and color diversity feature analysis test in no particular order. The method requires a subject to present her face before a camera, which can be the built-in or peripheral camera of e.g. a mobile communication device or a mobile computing device. The method also requires displaying to the subject certain instructions and the real-time video feedback of the subject face on a display screen, which can be the built-in or peripheral display screen of the mobile communication device or mobile computing device.

What is claimed is: 
     
       1. A face recognition method for personal identification and authentication comprising:
 capturing an input image of a subject to be identified and authenticated with a camera; 
 verifying, by a first processor, the identity of the subject by matching the input image against a database of pre-recorded face data records; 
 conducting, by a second processor, in no particular order, anti-spoofing tests on the input image including:
 a scan line detection test for detecting Moiré patterns created by an overlapping of digital grid from a digital media display and grid of the camera image sensor, wherein the input image is a spoof image if Moiré patterns are detected; 
 a specular reflection detection test for detecting one or more specular reflection features of a mirror or reflective surface from the input image, wherein the input image is a spoof image if one or more specular reflection features of a mirror or reflective surface are detected; and 
 a chromatic moment and color diversity feature analysis test; 
 
 wherein color diversity of the input image is analyzed to determine whether the input image is a spoof image; and 
 wherein the chromatic moment and color diversity analysis comprises:
 extracting chromatic features and color histogram features from the input image in both HSV and RGB spaces; and 
 classifying the extracted chromatic features and color histogram features to determine whether the input image is a spoof image. 
 
 
     
     
       2. The method of  claim 1 , wherein the scan line detection test comprises:
 a) applying band-pass filtering on the input image with a difference-of-Gaussians filter D(δ,k)=G(0, δ^2)−G(0, kδ^2) to generate a band-pass-filtered image, wherein G(0,δ^2) is a 2D-Gaussian function with zero mean and a standard deviation δ and k is the width of frequency band; 
 b) converting the band-pass-filtered image into frequency domain by discrete Fourier transformation and taking absolute values on the outputs; 
 c) applying thresholding on the band-pass-filtered image in frequency domain with a threshold T; 
 d) counting the number of pixels of the band-pass-filtered image in frequency domain with values higher than the threshold T and calculating the percentage, p, of the total number of pixels in the input image with values higher than the threshold T; and 
 e) determining whether high frequency peaks exist in the band-pass-filtered image in frequency domain, wherein the input image is a spoof image if p≤p min , wherein p min  is a pre-defined minimum value of percentage of the total number of pixels in the input image. 
 
     
     
       3. The method of  claim 2 , wherein the scan line detection test further comprises steps:
 increasing a standard deviation δ by an increment of Δ if p>p min ; and 
 repeating steps a) to e) if δ≤δ max , wherein δ max  is a pre-defined maximum value of δ. 
 
     
     
       4. The method of  claim 1 , wherein the specular reflection detection test comprises:
 extracting multi-dimensional specular reflection features from the input image; 
 discarding pixels of intensities outside of a pre-defined range; and 
 classifying the extracted specular reflection features to determine whether the input image is a spoof image. 
 
     
     
       5. The method of  claim 4 , wherein the pre-defined range is from one times a mean value of the pixels' intensities to five times the mean value. 
     
     
       6. The method of  claim 4 , wherein the extracted specular reflection features are classified with a support vector machine (SVM) based classifier trained with certain training sets. 
     
     
       7. The method of  claim 1 , wherein the extracted chromatic features and color histogram features are classified with a SVM based classifier trained with certain training sets. 
     
     
       8. A face recognition system for personal identification and authentication comprising:
 one or more computer processors, a camera, a storage media, and a display screen;
 wherein the face recognition system is configured to:
 capture an input image of a subject to be identified and authenticated with the camera; 
 verify identity of the subject by matching the input image against a database of pre-recorded face data records stored in the storage media; 
 
 conduct, in no particular order, anti-spoofing tests on the input image including:
 a scan line detection test for detecting Moiré patterns created by an overlapping of digital grid from a digital media display and grid of the camera image sensor, wherein the input image is a spoof image if Moiré patterns are detected; 
 a specular reflection detection test for detecting one or more specular reflection features of a mirror or reflective surface from the input image, wherein the input image is a spoof image image if one or more specular reflection features of a mirror or reflective surface are detected; and 
 a chromatic moment and color diversity feature analysis test, wherein the color diversity of the input image is analyzed to determine whether the input image is a spoof image. 
 
 
 
     
     
       9. The system of  claim 8 , wherein the scan line detection test comprises steps:
 a) applying band-pass filtering on the input image with a difference-of-Gaussians filter D(δ,k)=G(0, δ^2)−G(0, kδ^2) to generate a band-pass-filtered image, where G(0,δΛ2) is a 2D-Gaussian function with zero mean and a standard deviation δ and k is the width of frequency band; 
 b) converting the band-pass-filtered image into frequency domain by discrete Fourier transformation and taking absolute values on the outputs; 
 c) applying thresholding on the band-pass-filtered image in frequency domain with a threshold T; 
 d) counting the number of pixels of the band-pass-filtered image in frequency domain with values higher than the threshold T and calculating the percentage, p, of the total number of pixels in the input image with values higher than the threshold T; and 
 e) determining whether high frequency peaks exist in the filtered image in frequency domain, wherein the input image is a spoof image if p≤p min , where p min  is a pre-defined minimum value of percentage of the total number of pixels in the input image. 
 
     
     
       10. The system of  claim 9 , wherein the scan line detection test further comprises steps:
 increasing a standard deviation δ by an increment of Δ if p>p min ; and 
 repeating steps a) to e) if δ≤δ max , wherein δ max  is a pre-defined maximum value of δ. 
 
     
     
       11. The system of  claim 8 , wherein the specular reflection detection test comprises steps:
 extracting multi-dimensional specular reflection features from the input image; 
 discarding pixels of intensities outside of a pre-defined range; and 
 classifying the extracted specular reflection features to determine whether the input image is a spoof image. 
 
     
     
       12. The system of  claim 11 , wherein the pre-defined range is from one times a mean value of the pixels' intensities to five times the mean value. 
     
     
       13. The system of  claim 11 , wherein the extracted specular reflection features are classified with a support vector machine (SVM) based classifier trained with certain training sets. 
     
     
       14. The system of  claim 8 , wherein the extracted chromatic features and color histogram features are classified with a SVM based classifier trained with certain training sets. 
     
     
       15. The system of  claim 8 , wherein the system is a mobile communication or computing device. 
     
     
       16. The system of  claim 8 , wherein the system is a kiosk or a user terminal. 
     
     
       17. A face recognition method for personal identification and authentication comprising:
 capturing an input image of a subject to be identified and authenticated with a camera; 
 verifying, by a first processor, the identity of the subject by matching the input image against a database of pre-recorded face data records; 
 conducting, by a second processor, in no particular order, anti-spoofing tests on the input image including:
 a scan line detection test for detecting Moiré patterns created by an overlapping of digital grid from a digital media display and grid of the camera image sensor, wherein the input image is a spoof image if Moiré patterns are detected; 
 a specular reflection detection test for detecting one or more specular reflection features of a mirror or reflective surface from the input image, wherein the input image is a spoof image if one or more specular reflection features of a mirror or reflective surface are detected; and 
 a chromatic moment and color diversity feature analysis test; 
 
 extracting multi-dimensional specular reflection features from the input image; 
 discarding pixels of intensities outside of a pre-defined range; and 
 classifying the extracted specular reflection features to determine whether the input image is a spoof image; 
 wherein color diversity of the input image is analyzed to determine whether the input image is a spoof image.